Google's Frozen v2 Chip Would Embed Gemini Architecture in Silicon

Alphabet is developing a custom AI server chip, "Frozen v2," that bakes parts of Gemini's architecture directly into silicon, per The Information and Reuters on July 20, 2026. Slated for 2028, it could deliver 6–10x the inference efficiency of Google's current AI chips. The design trades general-purpose flexibility for gains only justified if Gemini's architecture stays stable through the chip's lifetime.